Ki67 immunohistochemistry (IHC), commonly used as a proliferation marker in breast cancer, has limited value for treatment decisions due to questionable analytical validity. The International Ki67 in Breast Cancer Working Group (IKWG) consensus meeting, held in October 2019, assessed the current evidence for Ki67 IHC analytical validity and clinical utility in breast cancer, including the series of scoring studies the IKWG conducted on centrally stained tissues. Consensus observations and recommendations are: 1) as for estrogen receptor and HER2 testing, preanalytical handling considerations are critical; 2) a standardized visual scoring method has been established and is recommended for adoption; 3) participation in and evaluation of quality assurance and quality control programs is recommended to maintain analytical validity; and 4) the IKWG accepted that Ki67 IHC as a prognostic marker in breast cancer has clinical validity but concluded that clinical utility is evident only for prognosis estimation in anatomically favorable estrogen receptor–positive and HER2-negative patients to identify those who do not need adjuvant chemotherapy. In this T1-2, N0-1 patient group, the IKWG consensus is that Ki67 5% or less, or 30% or more, can be used to estimate prognosis. In conclusion, analytical validity of Ki67 IHC can be reached with careful attention to preanalytical issues and calibrated standardized visual scoring. Currently, clinical utility of Ki67 IHC in breast cancer care remains limited to prognosis assessment in stage I or II breast cancer. Further development of automated scoring might help to overcome some current limitations.

Pathology is the cornerstone of cancer care. The need for accuracy in histopathologic diagnosis of cancer is increasing as personalized cancer therapy requires accurate biomarker assessment. The appearance of digital image analysis holds promise to improve both the volume and precision of histomorphological evaluation. Recently, machine learning, and particularly deep learning, has enabled rapid advances in computational pathology. The integration of machine learning into routine care will be a milestone for the healthcare sector in the next decade, and histopathology is right at the centre of this revolution. Examples of potential high‐value machine learning applications include both model‐based assessment of routine diagnostic features in pathology, and the ability to extract and identify novel features that provide insights into a disease. Recent groundbreaking results have demonstrated that applications of machine learning methods in pathology significantly improves metastases detection in lymph nodes, Ki67 scoring in breast cancer, Gleason grading in prostate cancer and tumour‐infiltrating lymphocyte (TIL) scoring in melanoma. Furthermore, deep learning models have also been demonstrated to be able to predict status of some molecular markers in lung, prostate, gastric and colorectal cancer based on standard HE slides. Moreover, prognostic (survival outcomes) deep neural network models based on digitized HE slides have been demonstrated in several diseases, including lung cancer, melanoma and glioma. In this review, we aim to present and summarize the latest developments in digital image analysis and in the application of artificial intelligence in diagnostic pathology.

Purpose: Because durable response to programmed cell death 1 (PD-1) inhibition is limited to a subset of melanoma patients, new predictive biomarkers could have clinical utility. We hypothesize that pretreatment tumor-infiltrating lymphocyte (TIL) profiles could be associated with response. Experimental Design: Pretreatment whole tissue sections from 94 melanoma patients treated with anti–PD-1 therapy were profiled by multiplex immunofluorescence to perform TIL quantification (CD4, CD8, CD20) and assess TIL activation (CD3, GZMB, Ki67). Two independent image analysis technologies were used: inForm (PerkinElmer) to determine cell counts, and AQUA to measure protein by quantitative immunofluorescence (QIF). TIL parameters by both methodologies were correlated with objective response or disease control rate (ORR/DCR) by RECIST 1.1 and survival outcome. Results: Pretreatment lymphocytic infiltration, by cell counts or QIF, was significantly higher in complete or partial response than in stable or progressive disease, particularly for CD8 (P < 0.0001). Neither TIL activation nor dormancy was associated with outcome. CD8 associations with progression-free survival (HR > 3) were independently significant in multivariable analyses and accounted for similar CD3 associations in anti–PD-1-treated patients. CD8 was not associated with melanoma prognosis in the absence of immunotherapy. Predictive performance of CD8 cell count (and QIF) had an area under the ROC curve above 0.75 (ORR/DCR), which reached 0.83 for ipilimumab plus nivolumab. Conclusions: Pretreatment lymphocytic infiltration is associated with anti–PD-1 response in metastatic melanoma. Quantitative TIL analysis has potential for application in digital precision immuno-oncology as an “indicative” companion diagnostic.

Purpose: Biomarkers for disease-specific survival (DSS) in early-stage melanoma are needed to select patients for adjuvant immunotherapy and accelerate clinical trial design. We present a pathology-based computational method using a deep neural network architecture for DSS prediction. Experimental Design: The model was trained on 108 patients from four institutions and tested on 104 patients from Yale School of Medicine (YSM, New Haven, CT). A receiver operating characteristic (ROC) curve was generated on the basis of vote aggregation of individual image sequences, an optimized cutoff was selected, and the computational model was tested on a third independent population of 51 patients from Geisinger Health Systems (GHS). Results: Area under the curve (AUC) in the YSM patients was 0.905 (P < 0.0001). AUC in the GHS patients was 0.880 (P < 0.0001). Using the cutoff selected in the YSM cohort, the computational model predicted DSS in the GHS cohort based on Kaplan–Meier (KM) analysis (P < 0.0001). Conclusions: The novel method presented is applicable to digital images, obviating the need for sample shipment and manipulation and representing a practical advance over current genetic and IHC-based methods.

Molecular profiling is central in cancer precision medicine but remains costly and is based on tumor average profiles. Morphologic patterns observable in histopathology sections from tumors are determined by the underlying molecular phenotype and therefore have the potential to be exploited for prediction of molecular phenotypes. We report here the first transcriptome-wide expression–morphology (EMO) analysis in breast cancer, where individual deep convolutional neural networks were optimized and validated for prediction of mRNA expression in 17,695 genes from hematoxylin and eosin–stained whole slide images. Predicted expressions in 9,334 (52.75%) genes were significantly associated with RNA sequencing estimates. We also demonstrated successful prediction of an mRNA-based proliferation score with established clinical value. The results were validated in independent internal and external test datasets. Predicted spatial intratumor variabilities in expression were validated through spatial transcriptomics profiling. These results suggest that EMO provides a cost-efficient and scalable approach to predict both tumor average and intratumor spatial expression from histopathology images.Significance: Transcriptome-wide expression morphology deep learning analysis enables prediction of mRNA expression and proliferation markers from routine histopathology whole slide images in breast cancer.